Shader Optimization – True Instruction Cost, Performance Tips // Tutorial

Support on Patreon:   / techartaid  
Get project files: https://techartaid.com/products/
Streaming weekly at   / techartaid  

Learn how to optimize material performance in HLSL and Unreal Engine. I use shader compiler to analyze GPU assembly instructions (which is easier than you think!). I explain the real speed of intrinsic math operations - how expensive is sin, cos, atan in GPU cycles. I also share tips about texture sampling cost, UVs. The process applies to all real time game engines that use HLSL language: Unity, Unreal Engine, Godot

Kostas Anagnostou - How to read shader assembly: https://interplayoflight.wordpress.co...

Fabian Giesen (Ryg) - A Trip Through Graphics Pipeline: https://fgiesen.wordpress.com/2011/07...

00:00 Intro
00:05 Shader assembly. How compilers work
00:08 Basic shader in HLSL
10:00 Cost of sin instruction (sine, cosine)
12:20 Test material in UE5. Are constants optimized?
15:46 Unreal instruction count is fake
20:19 How expensive is tan (tangent) on GPU
21:21 Be careful with material functions in Unreal
23:49 Do unconnected nodes contribute to instruction cost?
26:28 Can we beat the compiler?
31:21 Texture sampling. How mipmaps & cache improve speed
36:37 Free math trick (thanks to texture memory latency)
41:33 Scalar memory (SGPR) vs vector registers (VGPR)
44:20 Wait counter
47:13 Modifying UV coordinates can affect performance. Cache coherence
56:00 Packing grayscale into RGB channels in Substance Designer
59:26 Gradient mapping. Color curve atlas in Unreal Engine